基于近红外高光谱的梨叶片炭疽病与黑斑病识别  

Identifying Anthracnose and Black Spot of Pear Leaves on Near-infrared Hyperspectroscopy

作　　者：刘莉[1] 陶红燕 方静 郑文娟 王良龙 金秀[2] LIU Li;TAO Hongyan;FANG Jing;ZHENG Wenjuan;WANG Lianglong;JIN Xiu(College of Horticulture,Anhui Agricultural University,Hefei 230036,China;College of Information and Computer Science,Anhui Agricultural University,Hefei 230036,China)

机构地区：[1]安徽农业大学园艺学院,合肥230036 [2]安徽农业大学信息与计算机学院,合肥230036

出　　处：《农业机械学报》2022年第2期221-230,共10页Transactions of the Chinese Society for Agricultural Machinery

基　　金：财政部和农业农村部:国家现代农业产业技术体系建设专项(CARS2914);国家重点研发计划项目(2018YFD0201401);安徽省教育厅项目(KJ2019A0212)。

摘　　要：针对梨炭疽病和黑斑病发病症状很相似,难以区分,导致实际生产中不便对症施药的问题,以砀山酥梨叶片为研究对象,探究利用高光谱技术来识别梨叶片炭疽病与黑斑病的可行性。首先,运用高光谱成像系统采集砀山酥梨正常叶片、炭疽病叶片和黑斑病叶片的高光谱图像,提取图像的平均光谱反射率。采用多元散射校正法(Multiplicative scatter correction,MSC)、SavitzkyGolay卷积平滑法和标准正态变换法(Standard normal variate,SNV)分别对原始光谱数据进行预处理。然后,采用主成分分析算法(Principal component analysis,PCA)、连续投影算法(Successive projections algorithm,SPA)、无信息变量消除法(Uniformative variable elimination,UVE)、竞争性自适应重加权算法(Competitive adaptive reweighted sampling,CARS)、随机蛙跳算法(Shuffled frog leaping algorithm,SFLA)提取特征波长,分别获取了27、12、15、26、20条特征波长,并将其作为后期建模的输入变量。经对比发现,在各基于特征波长建立的支持向量机(SVM)分类识别模型以及BP神经网络分类识别模型中,SPASVM识别模型效果最佳,测试集准确率为93.25%,建模集准确率为94.80%。试验结果证明,利用高光谱技术能够有效识别砀山酥梨叶片的黑斑病与炭疽病。Pear anthracnose and pear black spot are serious diseases that occur during the growth of pears.The symptoms of these two diseases are very similar and it is difficult to distinguish,which leads to the inconvenience of prescribing the right medicine to these two kinds of leaves in actual production.In response to the status quo,taking‘Dangshan'pear leaves as the study object,the feasibility of using hyperspectral technology to identify anthracnose and black spot on pear leaves was explored.First of all,the hyperspectral imaging system was used to collect the hyperspectral images of the normal leaves,anthracnose leaves and black spot leaves of‘Dangshan'pear,and extract the average spectral reflectance of the images.The multiplicative scatter correction method(MSC),SavitzkyGolay convolution smoothing method and standard normal variate method(SNV)were used respectively to preprocess the original spectral data.Then the principal component analysis(PCA),successive projections algorithm(SPA),uniformative variable elimination(UVE),competitive adaptive reweighted sampling algorithm(CARS),and shuffled frog leaping algorithm(SFLA)were used to extract characteristic wavelengths,respectively,and totally 27,12,15,26 and 20 characteristic wavelengths were obtained,and using them as input variables for later modeling.After comparison,it was found that in the support vector machine(SVM)classification and recognition model based on characteristic wavelength and the BP neural network classification and recognition model based on characteristic wavelength,the SPASVM recognition model had the best effect during all models,the accuracy rate of the model's test set was 93.25%,and the accuracy rate of the model's modeling set was 94.80%.The test results proved that hyperspectral technology can effectively identify the black spot and anthracnose of‘Dangshan’pear leaves.

关 键 词：砀山酥梨 炭疽病 黑斑病 高光谱 特征波长 识别模型 

分 类 号：S436.612.1[农业科学—农业昆虫与害虫防治] S661.2[农业科学—植物保护]